Ancestrify field guide

Glossary

Name: Ancestrify
Availability: InStock

The vocabulary this field uses, defined plainly, and for the terms people routinely read too much into, what they do not mean.

Global25

Global25 and coordinate methods

Global25G25: A set of 25 numbers that locates one genome in a genetic reference space built from ancient and modern samples. Coordinates come from the independent Eurogenes Global25 service; comparing them means measuring distances and fitting mixtures inside that space.; What it is not. Not an ancestry test and not a set of ethnicity percentages. A coordinate is a position, and it only means something relative to the other positions you compare it against.
Coordinate row: One sample's 25 Global25 values, written as a single comma-separated line beginning with the sample's label. It is the input format every coordinate tool takes.
Scaled and unscaled coordinates: Two published forms of the same 25 values. The scaled form multiplies each dimension by that dimension's share of the variance, so the earlier dimensions carry more weight in a distance calculation; the unscaled form treats all 25 equally.; What it is not. Not interchangeable. Mixing scaled and unscaled rows in one comparison produces distances that mean nothing, and the arithmetic will not warn you, the numbers still come out looking reasonable.
Euclidean distance: The straight-line distance between two coordinates across all 25 dimensions: square the difference in each dimension, add them, take the square root. It is how closest-population rankings are produced.; What it is not. Not a measure of relatedness or of shared ancestors. Two groups can sit close together because they genuinely share ancestry, or because they are both mixtures that happen to average out to a similar point.
Admixture model: An estimate of your coordinate as a weighted mixture of chosen source populations, searched for by Monte-Carlo methods in the nMonte tradition and reported with a fit distance.; What it is not. Not a test. A coordinate fit cannot reject a model, it always returns percentages, including for a source panel containing nobody your ancestors met.
Fit distance: How far the fitted mixture's combined coordinate still sits from your own after the search finishes. Smaller means the mixture lands nearer your point.; What it is not. Not a p-value and not a confidence measure. A small fit distance says the arithmetic closed the gap, not that the historical story behind the sources is right.
Calculator: A curated, era-scoped set of source populations an admixture model is fitted against. Which calculator you choose shapes the answer more than the arithmetic does.
Era: A dated window that scopes which reference populations a comparison uses. Distances run across six: Late Bronze Age (3000 to 1200 BC), Pre-Classical Iron Age (1200 to 0 BC), Imperial Antiquity (0 to 600 AD), Middle Ages (600 to 1400 AD), Early Modern (1400 to 2000 AD) and the Modern Era (2000 AD onward).
PCAprincipal component analysis: A projection that flattens many dimensions into a plot you can read, placing samples so that the largest axes of variation become the axes of the chart.; What it is not. Not a map, and the axes are not ancestries. Distance on a two-dimensional scatter can hide separation that exists in the dimensions the plot dropped.
Population average: A coordinate built by taking the arithmetic mean of each dimension across the individuals in a group. Our reference set holds 1,535 curated populations averaged from 30,386 individual samples.; What it is not. Not a person. An average describes the centre of a group, and individuals scatter widely around it, which is why matching an average closely is not the same as matching anyone who lived.

qpAdm

Formal statistics and qpAdm

qpAdm: A formal admixture test that estimates a target's ancestry as proportions of chosen source populations, working from allele-frequency statistics and a set of outgroups. It returns a weight, a standard error and a z-score per source, plus a p-value for the model as a whole.; What it is not. Not a coordinate fit dressed up in different language. The defining difference is that qpAdm can reject a model outright, which no distance-based method can do.
ADMIXTOOLS 2: The R package these methods are implemented in, used in published ancient-DNA research. Our paid analysis runs its qpadm() function; the free Lab exposes f-statistics, qpWave, qpAdm and admixture-graph fitting against a reference panel.
p-value: The probability of seeing data at least this far from the proposed model if the model were true. A high value means the model is compatible with the data; a low one means it is not.; What it is not. Not the probability that the model is correct, and not a measure of how much ancestry came from anywhere. A model can pass with a comfortable p-value and still be historically wrong, because a different model can pass too.
Standard errorSE: The uncertainty around one estimated weight. It depends far more on how many markers your file carries after merging than on how long anyone searched for the model.; What it is not. Not something extra effort can shrink. Coverage sets the floor, which is why a sparse file cannot reach the tightest statistical bars no matter what is paid for it.
Z-score: How many standard errors a weight sits from zero. A source needs a large |Z| before its contribution can be called distinguishable from nothing at all.
95% confidence interval95% CI: The weight plus or minus 1.96 standard errors: the range the true proportion would plausibly fall in. Printed beside every source in a qpAdm report's model record, and worth reading before the percentage.; What it is not. Not a guarantee that the true value is inside it, and not a range of possible ancestries. An interval that reaches zero is a source the model cannot certify.
Chi-square and degrees of freedomchi-square, dof: The two numbers a qpAdm p-value is derived from: how far the observed matrix of f4 statistics sits from the closest matrix of the required rank, and how many independent contrasts the model had to fit. More outgroups mean more degrees of freedom and a harder test.
Nested modelpopdrop: A simpler model contained inside the published one, usually the same model with one source removed. The model record lists every such refit with its weights, p-value and feasibility; a source stays only because every model without it is rejected.
Jackknife blocks: How qpAdm computes its standard errors: the genome is cut into blocks of about five centimorgans, the model is refitted leaving each block out in turn, and the spread of those refits is the error. The block count is printed in the model record.
Model record: The complete output of the run a published qpAdm model came from, shown in the report's Reading view and downloadable as plain text: the fit block, every source with its panel label, sample count and confidence interval, the ordered outgroup set with counts, the nested-model table, the rank test and the tool's own warnings.; What it is not. Not a PDF and not a summary. It is the same figures the software produced, kept so that another analyst can check or reproduce the model.
Outgroupthe right set: A population used as a reference point rather than as a candidate ancestor. The outgroups are what give qpAdm its power to discriminate between models.; What it is not. Not a formality to be filled in. A right set that is too small, or too closely related to your sources, will accept almost any model you propose, the commonest way to produce a confident and meaningless result.
Left set and right set: The two halves of a qpAdm model: the left set is the target plus the candidate sources, the right set is the outgroups it is measured against.
Rotation: An automated strategy that moves populations between the left and right sets to search many models at once.; What it is not. Not what we publish from. Rotating qpAdm has a high false-discovery rate, so automated rotation was built here and then deliberately removed, every published model is composed and checked by hand.
f-statisticsf2, f3, f4, D: Summaries of shared genetic drift between sets of populations. They are the raw material qpAdm, qpWave and admixture graphs are all computed from.
AADRAllen Ancient DNA Resource: The public compilation of published ancient and modern genotypes that formal modelling here is run against, version 66, roughly 23,265 samples across roughly 6,015 distinct population labels.
qpWave: qpAdm's sibling program: it estimates how many independent streams of ancestry a set of populations carries relative to the outgroups, without naming the sources. The rank test printed in a qpAdm model record is a qpWave computation.; What it is not. Not a substitute for qpAdm. qpWave says how many waves of ancestry the data supports, never which populations delivered them or in what proportions.
Distal and proximal models: Two ways of choosing qpAdm sources for the same target. A distal model reaches for deep-time ancestries, hunter-gatherers, early farmers, steppe pastoralists; a proximal model uses populations close in time to the target, which reads more like history.; What it is not. Not rival methods, and neither is 'more correct'. The same target can pass both; each answers a different question, and a proximal source is itself a mixture of the distal ones.

Calculators

Admixture calculators and community tools

Admixture calculator: Any tool that decomposes a genome or a coordinate into percentage contributions from a fixed set of components or source populations. GEDmatch's project calculators fit chip genotypes against component allele frequencies; Global25-based calculators fit a coordinate against population averages.; What it is not. Not a test of any historical claim. Every calculator always returns percentages, including from source sets your ancestors never met, and different calculators legitimately disagree because their components and reference samples differ.
Knumber of components: The number of ancestral components a calculator was built with, Eurogenes K13 has thirteen, K15 has fifteen. Each component is a statistical construct derived from the reference samples used to train that calculator.; What it is not. Not a quality score. A higher K is a finer-grained decomposition of the same data, not a more accurate one, and components with geographic names are still constructs, not observed ancient populations.
Oracle: GEDmatch's population-fitting utility. After a calculator run, Oracle ranks reference populations by how closely their component percentages match yours, and Oracle-4 additionally fits you as small mixtures of them, both reported with a distance where smaller is closer.; What it is not. Not a relatedness measure. An Oracle distance compares calculator percentages, not DNA, two people can carry similar percentage profiles without sharing any recent ancestry.
Calculator effect: The bias that makes an admixture calculator describe the people whose samples trained it better than everyone else. The term dates to the 2012 dispute between the Eurogenes and Dodecad projects; the durable lesson is that a calculator is most reliable for populations well represented in its references.
ADMIXTUREthe software: The academic program behind the model-based clustering plots in population-genetics papers: it assigns each genome fractions of K statistically inferred components, estimated from all samples jointly, in unsupervised or supervised mode.; What it is not. Not what a hobbyist calculator runs on your upload. GEDmatch-style tools project one file onto component frequencies a past run produced, which is why a published K plot cannot be reproduced by uploading a single kit anywhere.
nMonte: An R script by Ger Huijbregts that fits a Global25 coordinate as a weighted mixture of source coordinates by Monte-Carlo search. Most community mixture tools, Vahaduo's admixture view included, descend from its approach, as does the seeded engine our Global25 reports run.; What it is not. Not a statistical test. Like every coordinate fit it always returns percentages and reports only a fit distance, never a p-value that could reject the model.

Segments

Segments, matches and lineages

IBDidentity by descent: A stretch of DNA two people share because both inherited it from a common ancestor.; What it is not. Not what a scan of ancient samples can establish. Demonstrating descent requires evidence a genotype comparison alone does not carry, which is why we describe our results as IBS.
IBSidentity by state: A stretch of DNA where two samples simply match, whether or not the match came from a shared ancestor. It is what a comparison against ancient genomes actually detects.; What it is not. Not proof of a family connection. A shared segment with an excavated individual is a genuine signal of shared ancestry in a population, never a claim that the person was your ancestor or relative.
Haplogroup: A branch on the single-line tree of either the Y chromosome (paternal) or the mitochondrial genome (maternal), defined by the mutations that mark it.; What it is not. Not an ethnicity, and not a summary of your ancestry. A haplogroup traces one thread, father's father's father, or mother's mother's mother, while almost all of your ancestry sits in the rest of your genome.
Clade: A haplogroup branch together with everything descending from it. A deeper clade is a more specific placement on the tree.
Subclade: A branch sitting below another on the haplogroup tree, a more specific placement within a broader lineage. R-M269 is a subclade of R1b; H1 is a subclade of H.
Haplotree: The whole branching tree of one uniparental lineage, every named paternal branch of the Y chromosome, or every named maternal branch of the mitochondrial genome, with the variants that define each one.
SNPsingle-nucleotide polymorphism: A single position in the genome where a base differs between people. SNPs are what genotyping chips read, and the ones on the Y chromosome are what define its haplogroup branches.
Terminal SNP: The deepest branch-defining variant a file can confirm, the end of the line your data supports. Positions further down the tree that the file never tested are reported as untested rather than assumed.; What it is not. Not necessarily the deepest branch you belong to. A chip tests a fixed set of positions, so a terminal call reflects the resolution of the data, not the limit of the tree.
Mitochondrial DNAmtDNA: The small genome inside the mitochondria, inherited by everyone from their mother alone. It records the maternal line and is read against the rCRS reference.; What it is not. Not carried only by women. Both sexes inherit mtDNA; only mothers pass it on.
rCRSrevised Cambridge Reference Sequence: The standard human mitochondrial genome that every mtDNA variant in the literature is stated against. A mutation “at 16189” means a difference from the rCRS at that position.
CentimorgancM: The unit of genetic distance used to size a shared segment, roughly, the length over which one crossover is expected per generation. Shared totals in centimorgans are how Ancient Matches reports the weight of a match.; What it is not. Not a physical length. Two segments of the same centimorgan value can span very different numbers of base pairs, because recombination is not uniform across the genome.
Coverage: How many usable markers a raw DNA file actually contains, and where they fall. It governs which analyses a file can support and how tight their standard errors can be.; What it is not. Not fixed per company. Files from the same provider differ by chip generation, and some products carry no Y-chromosome rows at all, so a file can work for one analysis and be unusable for another.
Genotype merge: Combining your markers with a reference panel so the two can be compared position by position. Only the positions present in both survive, which is why the merge, not the panel, sets the marker count a model is computed from.